Pricing
Why One AI Answer Costs $0.01—or $0.99
A dated, official-source comparison of five AI/search pricing models, followed by a fair worksheet for calculating cost per useful answer and a current Achla AI plan snapshot.
6 min read
What does one AI answer cost? Depending on the pricing page, the visible number can be less than a cent, a few cents, or $0.99.
That range is real, but it is not a clean price comparison. One vendor counts a search request. Another counts a chatbot response. A third charges once when a support conversation reaches an outcome. Some plans include a crawler and index; others package human handoff, service levels or channels around the answer.
The useful question is therefore not “Who charges less?” It is:
What event is being billed, what work surrounds it, and how many useful answers do I receive?
Answer first: the market does not have a standard “AI answer” unit. Normalize the denominator before comparing prices. A $0.99 support outcome may include several questions and workflow automation; a $0.007 chatbot response may be one turn inside a longer conversation.
A current price ladder—with the units left attached
Prices checked on official public pages on 1 August 2026:
- Site Search 360: Published entry or unit: $9/month for 2,000 searches; $0.004 per excess search · What the vendor counts: Classic search query · Comparison caveat: Returns search results; not a generated answer unit.
- ChatLab Basic: Published entry or unit: $18/month for 2,400 message credits; calculator shows about $0.007 per one-credit chatbot response · What the vendor counts: Chatbot response credits · Comparison caveat: Credit use varies by model and context; a conversation usually has multiple responses.
- SiteGPT Starter: Published entry or unit: $39/month billed annually; up to 4,000 messages · What the vendor counts: Messages · Comparison caveat: Its pricing calculator says volume varies by model and counts the user question plus AI reply.
- AddSearch AI Answers: Published entry or unit: From $8,400/year · What the vendor counts: Annual product package; volume not public on the page · Comparison caveat: Includes enterprise features such as governance, implementation support and a 99.9% SLA.
- Intercom Fin: Published entry or unit: $0.99 per outcome · What the vendor counts: Outcome, charged once per conversation · Comparison caveat: One conversation can contain several questions and support workflow steps.
These rows are deliberately not sorted from “cheapest” to “most expensive.” Doing so would imply equivalence that the vendors do not claim.
Site Search 360 is a classic site-search service. ChatLab and SiteGPT are knowledge-base chatbots. AddSearch sells generative answers inside an enterprise search package. Intercom Fin is a support agent priced around completed work. The interfaces may all contain an answer box, but the surrounding job is different.
Why the denominator changes the story
Suppose a visitor asks three follow-up questions.
A query-priced search service may count four searches. A chatbot may count the user’s messages, the bot’s responses, or model-weighted credits. An outcome-priced support agent may charge once for the whole conversation—if it meets the vendor’s definition of an outcome.
Now add what is included. A higher price may cover human handoff, multiple communication channels, workflows, analytics, governance, onboarding, a service-level agreement or an account team. A lower price may cover only retrieval, leaving the buyer to run the model, storage and front end.
This is not a reason to avoid comparison. It is a reason to compare the actual job.
For a documentation site, the job may be “answer a reader’s question and show the source.” For a support team, it may be “resolve an issue without an agent.” For ecommerce, it may be “find the right product with filters and merchandising rules.” Each deserves a different measure of value.
Calculate cost per useful answer
The headline unit is only an input. A site owner can calculate a more useful measure:
Cost per useful answer = total monthly cost ÷ answers that visitors judged useful and could verify
“Total monthly cost” should include the subscription, overage, separate model/API bills, infrastructure and meaningful operating time.
“Useful” should be defined before the test. A practical standard might require that the answer:
- addresses the visitor’s question;
- cites a source that actually supports it;
- preserves important conditions or exceptions;
- sends the visitor to the right next step;
- avoids a support contact when self-service is appropriate.
Also track the answer rate separately. If 1,000 questions enter the system and 700 receive answers, the plan ratio per query is not the same as cost per returned answer. A refusal may be the correct and safer result, but it still uses system resources and should remain visible in the analysis.
A four-week pilot with real questions is more informative than a vendor table. Sample answered, refused and weak-result cases. Have a person open the citations. Record which questions should have been answered but were not, and which should have been refused sooner.
Where Achla AI fits
Achla AI is designed for the narrower website job: crawl one site, retrieve its content, return a short cited answer, and keep the source links visible.
The current pricing snapshot is:
- Free: 100 searches and up to 100 pages.
- Start: $15 per month, 400 searches and up to 1,000 pages.
- Pro: $39 per month, 1,500 searches and up to 5,000 pages.
- Top-up: $20 for 1,000 additional searches.
Dividing price by included searches gives 3.75 cents per Start query and 2.6 cents per Pro query. A top-up block is 2 cents per query. Those are plan ratios, not claims about the provider’s internal model cost or the value of each answer.
The crawler, index and model calls are included, so the site owner does not need a separate LLM key. New subscriptions begin with a hard limit; automatic top-up is an explicit opt-in with user-editable frequency and billing-period limits.
Achla’s pricing is runtime-editable. Treat the numbers above as checked on 1 August 2026 and use the live pricing page as the canonical source.
The answer is not the only product
The price ladder becomes easier to understand when you separate four layers:
- Retrieval: finding relevant passages or pages.
- Generation: composing a direct response.
- Grounding and reviewability: connecting claims to inspectable sources and declining weak cases.
- Workflow: handoff, ticketing, channels, governance, analytics and service commitments.
A website may need the first three and nothing more. A support department may need all four. Paying for an unused workflow layer is wasteful; expecting website-search pricing to include enterprise support operations is equally unrealistic.
That is why absolute claims such as “the cheapest AI answer” or “the best value” do not survive scrutiny. Price is only meaningful for a defined unit, customer and job.
A buyer’s five-question comparison
Ask every vendor:
- What exactly consumes one unit?
- Does a refusal, retry or follow-up consume another unit?
- Which infrastructure and model costs are included?
- What happens at the limit: stop, throttle, top up, or auto-upgrade?
- Can I export enough logs to evaluate source quality and useful-answer rate?
Then choose the product whose unit matches your outcome.
For a small site that wants cited answers without building an AI stack, review Achla AI pricing and AI search for documentation. For an enterprise comparison, see Achla AI versus AddSearch. And if you want the visual version of this argument, watch the mapped price-ladder video below.
The number on the pricing card is not meaningless. It simply needs its unit attached.
The YouTube player is not loaded until you press play.
Watch on YouTube (Video: Why One AI Answer Costs $0.01—or $0.99)
